Methodology — juryrig

Models

Scikit-learn

1. HistGradientBoostingClassifier (primary)

Histogram-based gradient boosting from scikit-learn. Uses default hyperparameters (no custom tuning). Post-hoc sigmoid calibration applied via CalibratedClassifierCV.

Setting	Value
Hyperparameters	scikit-learn defaults
Calibration	sigmoid (Platt scaling) via `CalibratedClassifierCV`, cv=3
Preprocessing	`OrdinalEncoder` for categorical, `SimpleImputer` for missing
Random state	7

2. LogisticRegression (baseline)

Linear classifier from scikit-learn.

Setting	Value
Solver	`saga`
max_iter	2000
tol	1e-3
C	1.0 (default)
Preprocessing	`MaxAbsScaler` + `OneHotEncoder`, `SimpleImputer`
Random state	7

3. DummyClassifier (floor baseline)

strategy="prior" — predicts training-set class distribution for every sample. Establishes a performance floor.

4. Race-Association Regression Models

Separate logistic regressions used for adjusted race-association analysis (not for prediction):

Specification	Controls	Hyperparameters
Core-adjusted	County, charge severity, arrest type, age, gender, ethnicity	`LogisticRegression(max_iter=300, C=1.0, solver="saga", random_state=7)`
Charge-detail	Core controls + law section, charge weight, attempt flag	Same

Both use OneHotEncoder(drop="first") for categorical encoding.

Calibration

Calibration is applied to the HistGradientBoostingClassifier only.

Setting	Value
Method	Sigmoid (Platt scaling)
Cross-validation folds	3
Implementation	scikit-learn `CalibratedClassifierCV`

Calibration evaluation uses 10 quantile-based bins (pd.qcut) to compute bin-wise average predictions vs. average outcomes.

Evaluation Metrics

Metric	Role	Definition
Accuracy	Primary classification	Share of cases classified correctly
AUROC	Ranking quality	Area under the ROC curve
PR-AUC	Positive-class ranking	Area under the precision-recall curve
Brier Score	Calibration quality	Mean squared error of predicted probabilities (lower is better)

Model selection uses Brier score on the evaluation split. Subgroup metrics computed by Race, Ethnicity, Gender, age_bucket, and Region.

Feature Policy

19 features from OCA-STAT arraignment-time fields. The default configuration (include_all) includes Gender, Ethnicity, and Race.

Target

Category	Values
Positive	`Plea`, `Verdict-TFG`
Negative	`Dismissed`, `Dism-ACD`, `Verdict-ACQ`
Excluded	Pending, warrant, transfer, other, unknown states

Leakage-Excluded Fields

Outcome and post-arraignment fields excluded from features:

Docket Status
Disposition Type / Detail
Dismissal Reason
Most Severe Sentence
Fines / Fees / Surcharges Imposed

Feature List

Court TypeRegionDistrictCountyCourt Arresting AgencyArrest TypeArraign YearArraign Month Top Charge at ArraignmentSeverityWeightLaw Article.SectionAttempt FlagGenderEthnicity RaceArrest Age

Data Sources

Dataset	Source	Grain	Coverage
OCA-STAT	NY Office of Court Administration	One defendant-docket	2021–2025, monthly refresh
DCJS/OCA Supplemental Pretrial	NY Division of Criminal Justice Services / OCA	One criminal cycle	2019–2024 (Oct 2025 release)

Both datasets are public. Raw CSVs and processed outputs are not committed to the repository; only aggregate analysis results are published.

Libraries

Library	Purpose
`scikit-learn`	All classifiers, calibration, preprocessing, metrics
`pandas`	Data manipulation and feature engineering
`numpy`	Numerical operations
`pyarrow`	Parquet data storage
`matplotlib`	Figure generation (SVG output)

Model Card

Reference Run

Field	Value
Snapshot date	`2026-03-07`
Run ID	`20260307_224037`
Modeled rows	1,609,252
Test rows	271,701
Feature count	19
Maturity threshold	≥ 90% disposed share by cohort
Best model	HistGradientBoostingClassifier
Best AUROC	0.8644
Best Brier	0.1496